das¶

Delay-and-sum reconstruction, compounding over the sequence (transmit) dimension.

For each target, das computes a weighted sum of interpolated samples from all channels and sequence events. The receive delay from each channel to each target is computed internally from the channel and target positions. The offsets array provides the per-target transmit delay (one-way, in samples), and weights provides per-target apodization.

Signature¶

PythonMATLAB

ffdas.das(
    x, 
    srcpos, 
    dstpos, 
    offsets, 
    weights, 
    *,
    srcdir=None, 
    wavenum=0.0, 
    algorithm=Algorithm.DEFAULT, 
    use_fp16=False, 
    channels_trailing=False, 
    out=None, 
)

out = ffdas.das(x, srcpos, dstpos, offsets, weights)
out = ffdas.das(x, srcpos, dstpos, offsets, weights, srcdir, wavenum, algorithm, use_fp16, channels_trailing)

Parameters¶

Parameter	Python	MATLAB	Description
`x`	`([batch,] channels, seq, samples)`	`(samples, seq, channels[, batch])`	Channel data (RF or IQ).
`srcpos`	`(channels, 3)`	`(3, channels)`	Channel positions in sampling wavelengths.
`dstpos`	`(..., 3)`	`(3, ...)`	Target positions in sampling wavelengths.
`offsets`	`(seq, ...)`	`(..., seq)`	Per-target transmit time offsets in samples. Spatial dimensions must match `dstpos`.
`weights`	`(seq, ...)`	`(..., seq)`	Per-target apodization weights. Same shape as `offsets`.
`srcdir`	`(channels, 4)` or `None`	`(4, channels)` or `[]`	Directivity vectors. The first three components are the unit surface normal of each channel element; the fourth is the cosine of the sensitivity half-angle. Targets outside a channel's cone receive zero contribution from that channel.
`wavenum`	`float`	`single`	Wavenumber for phase rotation, typically `-2pifc/fs` for IQ data. Set to `0` to disable.
`algorithm`	`Algorithm`	`int32`	Algorithm variant. `DEFAULT` (0) selects automatically.
`use_fp16`	`bool`	`logical`	Use half-precision arithmetic.
`channels_trailing`	`bool`	`logical` (default `true`)	If `true`, channel data layout is `(samples, seq, channels[, batch])`. If `false`, `(samples, channels, seq[, batch])`.

Returns¶

Reconstructed output with the spatial dimensions of dstpos: ([batch,] ...) in Python, (...[, batch]) in MATLAB.

Example¶

PythonMATLAB

k = sampling_freq / sound_speed

output = ffdas.das(
    rf,               # (batch, 1024, 1, 512) — 1024 channels, 1 transmit, 512 samples
    channel_pos * k,  # (1024, 3)
    voxel_pos * k,    # (64, 64, 64, 3)
    offsets,          # (1, 64, 64, 64) — one transmit event
    weights,          # (1, 64, 64, 64)
    wavenum=-2 * math.pi * center_freq / sampling_freq,  # wavenumber in sampling wavelengths
)
# output: (batch, 64, 64, 64)

k = sampling_freq / sound_speed;

output = ffdas.das( ...
    rf, ...               % (512, 1, 1024, batch) — 512 samples, 1 transmit, 1024 channels
    channel_pos * k, ...  % (3, 1024)
    voxel_pos * k, ...    % (3, 64, 64, 64)
    offsets, ...          % (64, 64, 64, 1) — one transmit event
    weights, ...          % (64, 64, 64, 1)
    -2 * pi * center_freq / sampling_freq ...  % wavenumber in sampling wavelengths
);
% output: (64, 64, 64, batch)

das_sparse¶

Like das, but each target compounds over a per-target subset of \(n\) sequence events selected by sparse_indices, rather than all sequence events. This is useful for synthetic aperture setups where each target only uses a subset of transmissions.

Additional Parameters¶

Parameter	Python	MATLAB	Description
`sparse_indices`	`(n, ...)`, `int32`, 0-based	`(..., n)`, `int32`, 1-based	Indices into the sequence dimension of `x`. Each target compounds the \(n\) events given by these indices.

The offsets and weights arrays have shape (n, ...) in Python and (..., n) in MATLAB, matching sparse_indices rather than the full sequence dimension.

Signature¶

PythonMATLAB

ffdas.das_sparse(
    x, 
    srcpos, 
    dstpos,
    offsets, 
    weights, 
    sparse_indices,
    *,
    srcdir=None, 
    wavenum=0.0,
    algorithm=Algorithm.DEFAULT,
    use_fp16=False,
    channels_trailing=False,
    out=None, 
)

out = ffdas.das_sparse(x, srcpos, dstpos, offsets, weights, sparse_indices)
out = ffdas.das_sparse(x, srcpos, dstpos, offsets, weights, sparse_indices, srcdir, wavenum, algorithm, use_fp16, channels_trailing)